Shopping online is all well and good until you get ripped off. Maybe the “pristine vintage couch” you bought on Craigslist turned out to be a moth-eaten relic from IKEA. Maybe the “Bruce Springsteen concert tickets” you snagged for $30 each really were too good to be true. Or maybe you sent your credit card information to a vendor who seemed trustworthy, and now your bank is blowing up your phone with stern warning calls. We all know that shopping via classifieds and online marketplaces can be a little sketchy — but did you know that robots are already well on their way to making this particular experience a lot safer?
You’re already using AI for Spotify and Netflix, but you might not be aware that AI has recently entered the world of retail and e-Commerce, too. In a collaboration with IBM’s Watson, North Face uses natural language processing to help their customers find the “perfect jacket for your next adventure.” The subscription shopping service Stitch Fix uses visual search to analyze a customer’s Pinterest boards, giving them data about that customer’s tastes in clothing. Flipkart, India’s largest e-Commerce provider, recently launched an AI personal shopper called Mira, which asks customers a series of questions about what they want from a product to help refine search.
Shopping online at North Face is one thing, though, but trolling Craigslist for deals is another. Classified sites face unique challenges that other e-Commerce sites don’t, and AI just doesn’t work the same on one as it does on the other. A company selling its own products controls an item’s metadata from the beginning. When they add a blue sweater to the site, it’s already tagged with information about its color, size and other features. If a customer asks a chatbot to find them a blue sweater in size medium, the AI doesn’t have to work too hard to fulfill that request.
Advertisement
Classifieds, on the other hand, rely on unstructured, user-generated input. When Tom from Iowa sells his 1983 Aerosmith shirt on eBay, he’s not necessarily adding the standardized keywords that will make it easy to search for. An AI algorithm looking to understand classifieds listings must be able to understand the natural language of users (read: full sentences, not preset tags) or identify and categorize items based on pictures alone. As classifieds platforms don’t have much control over who posts, AI must also filter out spam listings jam-packed with keywords put up by shady sellers. It’s a big task for any algorithm to take on, no matter how well-trained it is.
Old Dog, New Tricks
So, as they make their way onto mobile, classifieds face a big challenge. (And yes, they are making their way onto mobile. Craigslist is facing stiff competition from mobile-first upstarts like LetGo and OfferUp, which is expected to process the sale of more than $20 billion in goods this year. Even tech giant Facebook has also gotten in on the game, launching its Marketplace app late last year.)
In the analog age, people searched newspaper classifieds for a few specific things: job listings, used cars, apartment rentals and sometimes even love. But as classifieds have migrated online, the quantity and variety of listings has skyrocketed. From pre-loved clothing to vintage collectibles, there’s almost nothing you can’t find on classifieds sites now, thanks to their millions of individual for-sale ads.
That shift presents a challenge for the industry. The classified ad model has endured for so long because of its simple, almost bare-bones format: pages of ads with a photo, price and description. But as the number of listings gets bigger, the classifieds also get harder to navigate through simple browsing and text search — and the number of shady-looking ads grow. (I recently saw an ad for a used dog kennel that screamed “this listing was posted by a serial killer.”)
As a result, many companies are turning to AI and machine learning to give their sites and apps a modern overhaul. It’s a move that won’t just reinvent the classified ad, but will also shape how tomorrow’s customers shop — and help them feel more secure as they do so.
Your New Robot Bodyguard
The process by which classifieds use AI is comparable to how the Internet filters out fake news. Just like classifieds listings, online news sites range from the high-quality and trustworthy (think New York Times) to the clickbaity and spammy. An AI-driven aggregator like Google News has to sort through all that content and make judgements as to which sources are legitimate and which are not (a 2013 patent filing revealed that Google News uses 13 different metrics to do so.) It also has to “read” and understand the content of articles so it can match them to user preferences and “personalize” the news.
When it comes to shifty or fraudulent users, AI will be able to use the full extent of data that classifieds apps and sites have in order to study user behavior, not just the end result (the listing or article). To detect fraud, for instance, our company uses machine learning to study the way “good” users behave on our platform, from the start of the listing process all the way to the final listing, and detect anomalies that might indicate a user is “bad.” This is in contrast to the way many apps and sites are currently using large teams of offline content moderators to recategorize incorrect listings, remove duplicates and remove spam.
Harder, Better, Faster, Stronger
What’s interesting about the application of AI to online classifieds is that it not only enhances backend performance, but can also greatly speed up and optimize the user journey. For instance, an AI algorithm can create accurate clusters of similar products (i.e. an iPhone 7 cluster) based on photos and some unstructured text, and analyze them based on which ones are sold vs. unsold to determine what the ideal market price for that product should be. If a platform can auto-suggest that ideal market price every time a seller posts a similar product, it saves users a lot of time they’d otherwise spend doing their own research or experimentation, removing a bit of friction from their user experience.
AI speeds up not just the listing process but the selling process as well. Overpricing is a common mistake that results in unsold items, which in turn creates a negative experience for sellers. Overcoming this with AI-driven price recommendations will improve the user experience as users can successfully and speedily sell more of their items in a consistent manner.
Of course, there are many other highly visible applications of AI to online classifieds, from unique interfaces for visual search to chatbots that recommend products. But we believe that the greatest benefit of AI will be a significant improvement of user experience through faster selling, improved discovery of relevant items, and of course helping out behind the scenes to protect listing quality. This is where AI in classifieds will truly shine in the next five years — and begin to transform how customers sell and buy online. Because no one wants to buy fake Bruce Springsteen tickets.
Siu Rui Quek is a co-founder of Carouselland is responsible for setting the overall strategy and direction for the company. Since the launch of Carousell in August 2012 together with co-founders Marcus Tan and Lucas Ngoo, Quek has been deeply involved in product strategy, international expansion and growth. He is passionate about technology and has been buying and selling online since he was 13. This has been a driving force behind Quek’s commitment to building a product-focused company focused on solving meaningful problems for people globally.
